* Move the output fusion swizzling before fusions
* Due to the tension between LDS consolidation and multibuffering,
remove the reuse-lds call before the output swizzle. As a consequence,
remove the "increasing total LDS usage" heuristic from output swizzle
enablemente, since it should probably be fine
* Fix an issue where fusion traversal wasn't working correctly,
resulting in insufficinetly vectorized writes to global memory despite
previous attempts to fix the issue
* Fix a test that wasn't using i8 LDS
* Update the packed arithmetic test to check for vectorized writes
* Add a guard in case the ExistingOps strictness is still letting LDS
writes into the output swizzle rewrite

* Move the output fusion swizzling before fusions
* Due to the tension between LDS consolidation and multibuffering,
remove the reuse-lds call before the output swizzle. As a consequence,
remove the "increasing total LDS usage" heuristic from output swizzle
enablemente, since it should probably be fine
* Fix an issue where fusion traversal wasn't working correctly,
resulting in insufficinetly vectorized writes to global memory despite
previous attempts to fix the issue
* Fix a test that wasn't using i8 LDS
* Update the packed arithmetic test to check for vectorized writes
* Add a guard in case the ExistingOps strictness is still letting LDS
writes into the output swizzle rewrite

…ert the enableApplicability change

…ve-output-swizzle